Installer: cleanup of the lazy Nix removal and install --clone

- The nix.mount drop-in is also removed when the uninstall fails or is
  interrupted (exit trap), and when daemon-reload failed after writing it.
- --dir with a trailing slash no longer puts the temporary clone inside
  the directory it replaces.
- The template's repository uses Nix's git when SteamOS has none, so the
  template marker exists there too (--clone needs it).
- A failed Nix uninstall during the reinstall of a broken Nix says so.
- clone_config: one branch per case; shorter comments, help and docs.
- Tests: an interrupted uninstall, --dir with a trailing slash; a
  redundant "used as it is" case dropped.

README.md

@@ -94,16 +94,15 @@ Your own config in a git repository: `--clone <git-url>` clones it into
 `~/nix-config` (`--dir <path>`, branch `--ref <branch>`) and installs it like
 `--flake <dir>`. URLs: `git@github.com:owner/repo`, `https://...` or
 `github:owner/repo`. Without a Home Manager configuration it first activates
-the template, whose git credential helpers then clone a private repository:
+the template, so a private repository clones with its git credential
-your GitHub CLI login (`~/.config/gh`, kept by `uninstall`) or
+helpers: your GitHub CLI login (`~/.config/gh`, kept by `uninstall`) or
-`~/.git-credentials`; git never asks for a password. If the clone fails, it
+`~/.git-credentials`, never a password prompt. If the clone fails, it offers
-offers `gh auth login` (browser or one-time code, works with 2FA); an SSH URL
+`gh auth login` (browser or one-time code, works with 2FA); an SSH URL uses
-uses your key instead. The clone then replaces the template. Keep the
+your key instead. Keep the template's `programs.git`/`programs.gh` lines
-template's `programs.git`/`programs.gh` lines (see [Usage](#usage)) in your
+(see [Usage](#usage)) in your config, or the helpers go with the switch to
-config, or the helpers go with the switch to it. Only a missing
+it. Only a missing `~/nix-config` or the unchanged template is replaced by
-`~/nix-config` or the unchanged template is replaced by the clone; anything
+the clone; anything else there is used as it is. So re-running continues
-else there is used as it is, without cloning. So re-running continues where
+where it stopped (after a failed switch it just switches again).
-it stopped (after a failed switch it just switches again).
 
 ```sh
 curl -fsSL https://steam-frame-nix.lhns.de | bash -s -- install --clone git@github.com:owner/my-config

docs/cleanup.md

@@ -46,27 +46,20 @@ desktop can't reach it with its own environment) and uses the installed
 `home-manager` if there is one, else Home Manager's `master`.
 
 `install --clone` takes SteamOS' `git`, else `nix run nixpkgs#git`, always
-with `GIT_TERMINAL_PROMPT=0`. Each step checks whether an earlier run did it
+with `GIT_TERMINAL_PROMPT=0`. "The unchanged template": `create_config`
-(Nix works: not installed). The target directory is cloned into only when it
+writes the hash of the configuration's files (without `.git`) to
-is missing, empty or the untouched template: `create_config` writes the hash
+`.git/steam-frame-nix-template`; it must still match, with at most one
-of the configuration's files (without `.git`) to
+commit. Any other directory is used as it is, `--ref` ignored. The template
-`.git/steam-frame-nix-template`, and the template counts as untouched while
+is created and activated only into an empty target without
-the files still match and there is at most one commit. Anything else (the
+`~/.config/home-manager`; with another configuration linked, git clones
-clone of an earlier run, a changed template, any other directory) is used
+with that one's helpers. The switch to an existing template is skipped when
-as it is: no clone, `--ref` ignored, straight to the switch.
+it is linked, Home Manager is installed and git has a credential helper.
-
+The clone goes to `.<dir>.clone.XXXXXX` next to the target (removed on
-Into an empty target, the template comes first, but only without a Home
+failure or exit), then replaces the template (checked unchanged again).
-Manager configuration (no `~/.config/home-manager`); with one, git clones
+`gh auth login --hostname github.com --git-protocol https`, with the Home
-with that configuration's helpers. The switch to the template is skipped
+Manager profile's `gh`, is offered only for `https://github.com/` URLs and
-when it is linked, Home Manager is installed and git has a credential
+with a terminal. Tests: section G of `modules/cleanup/check.nix` (a logging
-helper. The clone goes to `.<dir>.clone.XXXXXX` next to the target (removed
+git, a fake nix, `STEAM_FRAME_NIX_TTY` instead of `/dev/tty`).
-if it fails, the template stays), replaces the template (checked unchanged
-again), then Home Manager switches to it. A failed switch leaves the clone,
-so a rerun only switches. `gh auth login --hostname github.com
---git-protocol https`, with the Home Manager profile's `gh`, is offered only
-for `https://github.com/` URLs and with a terminal. Tests: section G of
-`modules/cleanup/check.nix` (a logging git, a fake nix,
-`STEAM_FRAME_NIX_TTY` instead of `/dev/tty`).
 
 `restart-check` (run by the session module's activation after
 `steamFrameUserServices`, and by `install` at its end) compares the
@@ -78,16 +71,16 @@ with the debugger drop-in, looks for a listener on port 8087 in
 `STEAM_FRAME_NIX_CGROUP` (section F of `modules/cleanup/check.nix`).
 
 `uninstall` removes Nix with nix-installer (from a root-owned copy in
-`/tmp`), whose `systemctl stop nix.mount` fails while a process uses `/nix`,
+`/tmp`), whose `systemctl stop nix.mount` fails while a process uses `/nix`;
-and in the Steam session some always do until logout (e.g. Steam and
+in the Steam session some always do until logout (e.g. Steam and
-xdg-desktop-portal, once they mapped a file from the store). So first a
+xdg-desktop-portal, once they mapped a file from the store). So a runtime
-runtime drop-in, `/run/systemd/system/nix.mount.d/50-steam-frame-nix-lazy-unmount.conf`
+drop-in, `/run/systemd/system/nix.mount.d/50-steam-frame-nix-lazy-unmount.conf`
 with `LazyUnmount=yes`, makes that stop detach `/nix` (`umount -l`): the
 programs keep their open files until they exit. The drop-in is removed
-afterwards. Before that, for information, it lists the processes whose
+afterwards, also on failure or Ctrl+C. Listed first, for information: the
-`exe`, `cwd`, `root` or an `fd` links into `/nix`, or whose `maps` names a
+processes whose `exe`, `cwd`, `root` or an `fd` links into `/nix`, or whose
-file there (only the user's own are readable), except the script itself,
+`maps` names a file there (only the user's own are readable), except the
-its subshells and its process group (the `curl | bash` pipeline). Tests:
+script itself, its subshells and its process group (the `curl | bash`
-section H of `modules/cleanup/check.nix` (`STEAM_FRAME_NIX_PROC` as a fake
+pipeline). Tests: section H of `modules/cleanup/check.nix`
-`/proc`, `STEAM_FRAME_NIX_SYSTEM_RUNTIME` for the drop-in, a logging
+(`STEAM_FRAME_NIX_PROC` as a fake `/proc`, `STEAM_FRAME_NIX_SYSTEM_RUNTIME`
-nix-installer).
+for the drop-in, a logging nix-installer).

install.sh

@@ -7,7 +7,7 @@
 #   curl -fsSL https://steam-frame-nix.lhns.de | bash -s -- cleanup --all
 #
 # Run `install.sh --help` for details. Works from a file or piped into bash
-# (prompts read from /dev/tty; tests: STEAM_FRAME_NIX_TTY).
+# (prompts read from /dev/tty).
 set -euo pipefail
 
 TEMPLATE="${STEAM_FRAME_NIX_TEMPLATE:-github:lhns/steam-frame-nix}"
@@ -28,13 +28,14 @@ USER_NAME="$(id -un)"
 USER_ID="$(id -u)"
 OUTER_RUNTIME_DIR="${STEAM_FRAME_NIX_RUNTIME_DIR:-/run/user/$USER_ID}"
 SFN_PROC="${STEAM_FRAME_NIX_PROC:-/proc}"   # tests: a fake /proc
-SFN_TTY="${STEAM_FRAME_NIX_TTY:-/dev/tty}"
+SFN_TTY="${STEAM_FRAME_NIX_TTY:-/dev/tty}"  # tests: a file
-# Runtime drop-in for the uninstall (tests: another directory).
+# uninstall_nix: nix.mount detaches lazily (tests: another directory)
 NIX_MOUNT_DROPIN="${STEAM_FRAME_NIX_SYSTEM_RUNTIME:-/run/systemd/system}/nix.mount.d/50-steam-frame-nix-lazy-unmount.conf"
 
 ASSUME_YES=0
 RO_RELOCK=0
-CLONE_TMP=''
+CLONE_TMP=''      # an unfinished --clone, removed on exit
+NIX_DROPIN=0      # NIX_MOUNT_DROPIN written, removed on exit
 
 # --- output -----------------------------------------------------------------
 
@@ -63,12 +64,11 @@ Commands:
                              --dir (default ~/nix-config) on branch --ref,
                              then used like --flake <dir>. Without a Home
                              Manager configuration it activates the template
-                             first, so git clones with its credential helpers
+                             first, for git's credential helpers (gh,
-                             (gh, ~/.git-credentials); if that fails it offers
+                             ~/.git-credentials), and offers 'gh auth login'
-                             'gh auth login'. Only a missing directory or the
+                             if the clone fails. Only a missing directory or
-                             unchanged template is replaced by the clone;
+                             the unchanged template is replaced by the clone;
-                             anything else there is used as it is (a rerun
+                             anything else there is used as it is
-                             after a failed switch just switches)
         otherwise            ~/.config/home-manager
         neither exists       a new config in ~/nix-config from the
                              steam-frame-nix template, linked to
@@ -174,7 +174,8 @@ ro_relock() {
 }
 
 on_exit() {
-  [[ -z $CLONE_TMP ]] || rm -rf -- "$CLONE_TMP"   # an unfinished --clone
+  [[ -z $CLONE_TMP ]] || rm -rf -- "$CLONE_TMP"
+  nix_dropin_remove
   ro_relock
 }
 trap on_exit EXIT
@@ -248,8 +249,16 @@ install_nix() {
 
 nix_mounted() { mountpoint -q /nix 2>/dev/null; }
 
+nix_dropin_remove() {
+  (( NIX_DROPIN )) || return 0
+  NIX_DROPIN=0
+  sudo rm -f "$NIX_MOUNT_DROPIN"
+  sudo rmdir "${NIX_MOUNT_DROPIN%/*}" 2>/dev/null || true
+  sudo systemctl daemon-reload || true
+}
+
 uninstall_nix() {
-  local rc=0 bin='' lazy=0 users=()
+  local rc=0 bin='' users=()
   need_sudo
   ro_unlock
   step "Uninstalling Nix"
@@ -268,22 +277,16 @@ uninstall_nix() {
   # `systemctl stop nix.mount` would fail. With LazyUnmount= that stop
   # detaches /nix; the programs keep their open files until they exit.
   if nix_mounted; then
-    if sudo mkdir -p "${NIX_MOUNT_DROPIN%/*}" \
+    NIX_DROPIN=1
-       && printf '%s\n' '# steam-frame-nix: removing Nix detaches /nix while programs use it.' \
+    sudo mkdir -p "${NIX_MOUNT_DROPIN%/*}" \
-            '[Mount]' 'LazyUnmount=yes' | sudo tee "$NIX_MOUNT_DROPIN" >/dev/null \
+      && printf '%s\n' '# steam-frame-nix: removing Nix detaches /nix while programs use it.' \
-       && sudo systemctl daemon-reload; then
+           '[Mount]' 'LazyUnmount=yes' | sudo tee "$NIX_MOUNT_DROPIN" >/dev/null \
-      lazy=1
+      && sudo systemctl daemon-reload \
-    else
+      || warn "could not add $NIX_MOUNT_DROPIN; removing Nix fails if a program still uses /nix"
-      warn "could not add $NIX_MOUNT_DROPIN; removing Nix fails if a program still uses /nix"
-    fi
   fi
   sudo "$bin" uninstall --no-confirm || rc=$?
   [[ $bin == "$NIX_INSTALLER_BIN" ]] || sudo rm -f "$bin"
-  if (( lazy )); then
+  nix_dropin_remove
-    sudo rm -f "$NIX_MOUNT_DROPIN"
-    sudo rmdir "${NIX_MOUNT_DROPIN%/*}" 2>/dev/null || true
-    sudo systemctl daemon-reload || true
-  fi
   hash -r   # commands found in /nix are gone
   ro_relock
   return "$rc"
@@ -299,7 +302,7 @@ ensure_nix() {
       warn "Nix is installed ($NIX_RECEIPT) but not working"
       confirm "Uninstall and reinstall Nix?" || die "aborted"
       [[ -x $NIX_INSTALLER_BIN ]] || die "$NIX_INSTALLER_BIN is missing; can't uninstall the broken Nix"
-      uninstall_nix
+      uninstall_nix || die "the Nix uninstall failed (see above)"
     elif [[ -d /nix ]] && [[ -n $(ls -A /nix 2>/dev/null) ]]; then
       die "/nix exists but was not installed by nix-installer (no $NIX_RECEIPT); remove it or fix that Nix first"
     fi
@@ -326,23 +329,20 @@ set_let() { # file name value
   sed -i "s|^\( *$name = \)\"[^\"]*\";|\1\"$value\";|" "$file"
 }
 
-# Marker of a configuration create_config made: the hash of its files, in
+# create_config's marker in .git: the hash of the files it wrote, so
-# .git (never a file of the configuration). `install --clone` replaces
+# `install --clone` replaces the template only while it is unchanged.
-# such a configuration while it is unchanged (template_untouched).
 TEMPLATE_MARK=steam-frame-nix-template
 
-# Hash of a configuration's files (without .git).
+config_hash() { # dir: its files without .git
-config_hash() {
   (cd "$1" && find . -path ./.git -prune -o ! -type d -print0 | LC_ALL=C sort -z \
      | xargs -0 -r sha256sum | sha256sum) | cut -d' ' -f1
 }
 
-# Whether <dir> is the template as create_config made it: the marker
+# The marker matches the files, at most one commit.
-# matches its files, at most one commit.
 template_untouched() { # dir
   local m=$1/.git/$TEMPLATE_MARK n
   [[ -f $m ]] || return 1
-  n="$(git -C "$1" rev-list --count HEAD 2>/dev/null || echo 0)"
+  n="$(git_any -C "$1" rev-list --count HEAD 2>/dev/null || echo 0)"
   [[ $(<"$m") == "$(config_hash "$1")" ]] && (( n <= 1 ))
 }
 
@@ -370,16 +370,11 @@ create_config() { # dir [question]
   info "user $USER_NAME, home $HOME, system $system"
 
   # Flakes in a git repository only see tracked files.
-  if command -v git >/dev/null 2>&1; then
+  git_any -C "$dir" init -q
-    git -C "$dir" init -q
+  git_any -C "$dir" add -A
-    git -C "$dir" add -A
+  nix flake lock "$dir"
-    nix flake lock "$dir"
+  git_any -C "$dir" add flake.lock
-    git -C "$dir" add flake.lock
+  config_hash "$dir" >"$dir/.git/$TEMPLATE_MARK"
-    config_hash "$dir" >"$dir/.git/$TEMPLATE_MARK"
-  else
-    warn "git not found; $dir is used as a plain directory"
-    nix flake lock "$dir"
-  fi
 
   mkdir -p "${HM_CONFIG_LINK%/*}"
   ln -s "$dir" "$HM_CONFIG_LINK"
@@ -425,15 +420,11 @@ gh_login() { # url
   "$gh" auth login --hostname github.com --git-protocol https <"$SFN_TTY"
 }
 
-# --clone: <dir> missing (or empty) or the untouched template gets the clone
+# --clone, resumable: a missing (or empty) <dir> or the untouched template
-# of <url>; anything else in <dir> is used as it is (also the clone of an
+# there gets the clone of <url>; anything else is used as it is (e.g. the
-# earlier run whose switch failed). Each step checks whether an earlier run
+# clone of an earlier run whose switch failed). Without an active Home
-# did it already:
+# Manager configuration the template is activated first, for git's
-#   - without a Home Manager configuration (or with the template in <dir>
+# credential helpers.
-#     linked), the template is created in <dir> and activated, so git has
-#     the credential helpers (skipped if they are there);
-#   - <url> is cloned next to <dir> (on failure the template stays active
-#     in <dir>); the template is removed and the clone takes its place.
 clone_config() { # url dir ref
   local url dir=$2 ref=$3 tmpl=0 parent link=''
   url="$(clone_url "$1")"
@@ -445,21 +436,21 @@ clone_config() { # url dir ref
     tmpl=1
   fi
 
-  # The template, for git's credential helpers.
   if [[ -L $HM_CONFIG_LINK || -e $HM_CONFIG_LINK ]]; then
     link="$(readlink -f "$HM_CONFIG_LINK" || true)"
   fi
-  if (( ! tmpl )) && [[ -z $link ]]; then
+  if [[ -z $link ]]; then
-    create_config "$dir" "No Home Manager configuration found. First activate the steam-frame-nix template from $dir (git's credential helpers, e.g. 'gh auth login'), then replace it with the clone of $url?"
+    if (( tmpl )); then   # an earlier run's template; the link was removed
-    tmpl=1
-    hm_activate "$dir#$USER_NAME" || die "the switch to the template failed (see above); run install --clone again to continue"
-  elif (( tmpl )) && [[ -z $link || $link -ef $dir ]]; then
-    if [[ -z $link ]]; then
       mkdir -p "${HM_CONFIG_LINK%/*}"
       ln -s "$dir" "$HM_CONFIG_LINK"
       info "linked $HM_CONFIG_LINK -> $dir"
+    else
+      create_config "$dir" "No Home Manager configuration found. Activate the steam-frame-nix template in $dir first (for git's credentials), then replace it with the clone of $url?"
+      tmpl=1
     fi
-    if [[ -n $link ]] && hm_installed && git_has_helpers; then
+    hm_activate "$dir#$USER_NAME" || die "the switch to the template failed (see above); run install --clone again to continue"
+  elif (( tmpl )) && [[ $link -ef $dir ]]; then
+    if hm_installed && git_has_helpers; then
       step "The template in $dir is active with git's credential helpers (skipped)"
     else
       hm_activate "$dir#$USER_NAME" || die "the switch to the template failed (see above); run install --clone again to continue"
@@ -516,6 +507,8 @@ cmd_install() {
     clone_url "$clone" >/dev/null
     clone_dir=${clone_dir:-$DEFAULT_CONFIG_DIR}
     [[ $clone_dir == /* ]] || clone_dir=$PWD/$clone_dir
+    clone_dir=${clone_dir%"${clone_dir##*[!/]}"}   # no trailing /
+    [[ -n ${clone_dir%/*} ]] || die "--dir: '$clone_dir' is not below a directory"
   elif [[ -n $clone_dir || -n $clone_ref ]]; then
     die "install: --dir and --ref need --clone"
   fi

modules/cleanup/check.nix

@@ -448,20 +448,14 @@ pkgs.runCommand "cleanup-check" { nativeBuildInputs = [ cleanup pkgs.jq pkgs.git
   has "$res" "hm switch --flake $HOME/nix-config -b"
   [ "$($GIT -C $HOME/nix-config config --get remote.origin.url)" = git@github.com:owner/config ] || fail "existing directory changed"
 
-  # --dir (relative) and --ref; an active configuration: no template, the
+  # --dir (relative, trailing /) and --ref; an active configuration: no
-  # existing link stays
+  # template, the existing link stays
-  res=$(cd $HOME && inst --clone github:owner/config --dir cfg --ref dev)
+  res=$(cd $HOME && inst --clone github:owner/config --dir cfg/ --ref dev)
   grep -qF "clone --branch dev -- https://github.com/owner/config.git $HOME/.cfg.clone." $LOG || fail "ref: $(cat $LOG)"
   ! switched_to_template || fail "template with a configuration"
   clone_ok $HOME/cfg dev
   has "$res" "hm switch --flake $HOME/cfg -b"
   [ "$(readlink $HOME/.config/home-manager)" = $HOME/nix-config ] || fail "link replaced"
-  # not a configuration: not cloned over, kept (and no flake to switch to)
-  mkdir $HOME/plain; echo x > $HOME/plain/x
-  res=$(inst --clone github:owner/config --dir $HOME/plain 2>&1) && fail "plain dir used"
-  has "$res" "using it as it is (not cloning)"
-  has "$res" "has no flake.nix"
-  there $HOME/plain/x
   # a failed clone: the hint, nothing left
   res=$(inst --clone https://example.org/missing.git --dir $HOME/m 2>&1) && fail "missing repo"
   has "$res" "gh auth login"
@@ -542,7 +536,7 @@ pkgs.runCommand "cleanup-check" { nativeBuildInputs = [ cleanup pkgs.jq pkgs.git
   # nix-installer: logs its arguments and the lazy-unmount drop-in it sees
   D=$root/run-system/nix.mount.d/50-steam-frame-nix-lazy-unmount.conf
   export D LOG=$root/log RC=0
-  printf '#!%s\n[ "$1" = --version ] && exit 0\necho "nix-installer $*" >> $LOG\ncat $D >> $LOG 2>/dev/null || echo "no drop-in" >> $LOG\nexit $RC\n' \
+  printf '#!%s\n[ "$1" = --version ] && exit 0\necho "nix-installer $*" >> $LOG\ncat $D >> $LOG 2>/dev/null || echo "no drop-in" >> $LOG\n[ $RC != term ] || { kill -TERM $PPID; exit 1; }\nexit $RC\n' \
     "$(command -v bash)" > $root/nix-installer
   chmod +x $root/nix-installer
   uninst() { # mounted: 1|0
@@ -575,6 +569,10 @@ pkgs.runCommand "cleanup-check" { nativeBuildInputs = [ cleanup pkgs.jq pkgs.git
   res=$(RC=1 uninst 1) && fail "uninstall succeeded: $res"
   has "$res" "its uninstaller failed. Reboot, then run uninstall again"
   gone $D $HOME/.config/home-manager
+  # interrupted (TERM): the drop-in goes too
+  res=$(RC=term uninst 1) && fail "interrupted uninstall succeeded: $res"
+  has "$(cat $LOG)" "LazyUnmount=yes"
+  gone $D
   SH
   echo "H ok"
   touch $out

template/home.nix

@@ -10,9 +10,8 @@
   programs.home-manager.enable = true; # the `home-manager` command
 
   # git credentials for HTTPS (`install.sh install --clone` uses them):
-  # GitHub through the GitHub CLI (log in once: gh auth login, works with
+  # github.com through the GitHub CLI (gh auth login), other hosts from
-  # 2FA), other hosts from ~/.git-credentials. git itself is SteamOS's;
+  # ~/.git-credentials. git is SteamOS's; Home Manager writes its config.
-  # Home Manager only writes its config. Remove to manage git yourself.
   programs.git = {
     enable = true;
     package = null;